ECONOMIC STUDIES

DEPARTMENT OF ECONOMICS SCHOOL OF BUSINESS, ECONOMICS AND LAW

UNIVERSITY OF GOTHENBURG 220

________________________

Determinants of Health Capital at Birth: Evidence from Policy Interventions

Simona Bejenariu

ISBN 978-91-85169-95-5 (printed) ISBN 978-91-85169-96-2 (pdf) ISSN 1651-4289 print ISSN 1651-4297 online

Printed in Sweden,

Ineko 2015

To Valentin and our families.

i

ii

Acknowledgments

During the last five years, I thought many times about the moment I would writethese Acknowledgments. I thought it would be marked by a feeling of accomplish-ment and joy. But it is a time of reflection, of looking back at all the opportunitiesI was given, and most of all of being truly grateful for all the help I received togetting here. Being a PhD student meant more than just working towards a degree,it was an extremely valuable life experience, that I am grateful to have received.Standing at this milestone, I am both happy and sad it is so close to finishing.

I am deeply grateful to my supervisors, Andreea Mitrut and Randi Hjalmarsson.Andreea became more than supervisor to me, she is a mentor and a very closefriend. She generously shared her knowledge and experience, and taught me mostof what I know about research. Her constant encouragement, optimism and warmthgave me a feeling of comfort and trust that meant so much to me during these yearsand helped me overcome the occasional hold-backs. I am grateful to her for somany things that I cannot even express them all. I am also extremely grateful toRandi, who motivated me to strive and do better, guided and prepared me for mynext steps after finishing this PhD. She read every version of my single authoredpaper and never seemed tired of giving me excellent feedback and suggestions. Ifelt like I was learning to ride a bike without the training wheels, while knowingthat I am watched over and guided by someone who cared. I am grateful for all hersupport. I have been extremely fortunate to have them both as role models!

This thesis benefited greatly from the comments of Peter Nilsson from Stock-holm University and Hans Grnqvist from Uppsala University who were my oppo-nents at the Licentiate and Final seminars. Their insightful comments helped meimprove the papers a great deal.

I would like to thank all my colleagues at the Department of Economics atthe University of Gothenburg for the stimulating and supportive environment they

iii

created. I am especially grateful to Yonas Alem, Kristian Bolin, Dick Durevall,Lennart Flood, Lennart Hjalmarsson, Olof Johansson-Stenman, Per Krusell, Kata-rina Nordblom, Ola Olsson, Amrish Patel, Oleg Shchetnin, Eyerusalem Siba, JohanStennek, Michele Valsecchi and Conny Wollbrant, who offered excellent coursesand mentoring. Mns Sderbom was an inspiring teacher, who gave me timeand again valuable academic advice. Simon Felgendreher, Verena Kurz, Efi Kyri-akopoulou and Andrea Martinangeli became good friends of mine who made com-ing to work (and after work!) a pleasure! I also thank sa Adin, sa Lfgren, Eva-Lena Neth-Johansson, Selma Oliveira, Ann-Christin Rtri Nystrm and JeanetteSaldjoughi for their great administrative support.

My peers played an important part in this experience. Sied Hassen, MarcelaJaime, Reda Moursli, Diem Van Nguyen, Emil Persson, Remidius Ruhinduka,Joakim Ruist, Anja Tolonen, Hang Yin and Xiao-Bing Zhang - thank you all foryour friendship! I will always cherish the memory of you showing up for my sur-prise 25th birthday! A very special thanks goes to Oana Borcan, one of my dearestfriends ever. We shared the joys and sorrows of this PhD, but also recipes, the latestnews from home, memories and hopes. Thank you for being there for me!

Luckily for me, as I have never been much of an "workaholic", I encounteredmany great people that became our very good friends. Lavinia, Costel, Vlad, Elena,Cri, Cesar, Diana, Ruxi, Alin, Irina, Mihai B., Mihai C., Daniel, Dani, Claudia,Laurentiu, Thomas, Ioannis, Vasiliki, Iosif and Maria, thanks to you I have so manynice memories when I look back at all these years!

I dedicate this thesis to my family. My mother, father and brother have alwaysbeen close to my heart. This thesis meant that we missed birthdays and other im-portant moments in each others lives, but their love shone through and the distancefaded away. The hundreds of hours of Skype calls were of immense help, and Icannot express my gratitude for their constant support. Va multumesc! Above all, Iowe what Ive accomplished to Valentin, my husband-to-be. He has been my great-est support and his unconditional love made me feel I could never fail. Nearly sixyears ago we were two kids who packed their bags in a car and drove off together.We didnt know where our journey would lead, we still dont, but I am sure that byeach others side, the road ahead is bright and we will make the best of all that is tocome. ti multumesc pentru tot, te iubesc!

Simona BejenariuGteborg, April 2015

iv

Contents

Acknowledgments iii

1 Introduction 11.1 Early childhood and human capital formation . . . . . . . . . . . 1

1.2 Romanian context . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Summary of the thesis . . . . . . . . . . . . . . . . . . . . . . . 4

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Austerity Measures and Infant Health. Lessons from anUnexpected Wage Cut 112.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2 Background and data . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3 Identification and main results . . . . . . . . . . . . . . . . . . . 22

2.4 Potential mechanisms . . . . . . . . . . . . . . . . . . . . . . . . 28

2.5 Further evidence of selection in utero . . . . . . . . . . . . . . . 33

2.6 Further sensitivity checks . . . . . . . . . . . . . . . . . . . . . . 36

2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3 Bridging the Gap for Roma Women. The Effects ofa Health Mediation Program on Roma Prenatal Care andChild Health at Birth 713.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2 The Roma Health Mediators Program . . . . . . . . . . . . . . . 76

3.3 Data and methodology . . . . . . . . . . . . . . . . . . . . . . . 78

3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

v

3.5 Robustness tests . . . . . . . . . . . . . . . . . . . . . . . . . . 893.6 Further evidence using survey data . . . . . . . . . . . . . . . . . 923.7 Potential Mechanisms and Discussion . . . . . . . . . . . . . . . 943.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4 The Effects of Financial Incentives on Fertilityand Early Investments in Child Health 1314.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1324.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1354.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1474.4 Identification strategy . . . . . . . . . . . . . . . . . . . . . . . . 1534.5 Main results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1574.6 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . 1644.7 Heterogeneity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1714.8 Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . 172Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

vi

Chapter 1

Introduction

1.1 Early childhood and human capital formation

Childhood shows the man as morning shows the day wrote John Milton overthree centuries ago. Indeed, mounting evidence in economics and other domainssuggests that early childhood environment can have persistent impacts on numer-ous later life outcomes, primarily mediated through the accumulation of humancapital ( [Currie and Almond, 2011]). One of the most salient components of thisenvironment is health status during infancy and childhood that plays a crucial rolein the formation of human capital and explains a significant share of the varia-tion in several important non-health adult outcomes. Birth weight, a widely usedmeasure of the childs endowment at birth, is a strong predictor not only of healthstatus in adulthood, but also educational attainment, labor market participation andincome, whereas low birth weight, the condition of being born with a weight be-low 2,500 grams, has significant negative long-term effects on the aforementionedoutcomes ( [Currie and Hyson, 1999, Black et al., 2007]). Moreover, there is con-clusive evidence of a strong correlation between the birth weight of mothers andthe birth weight of their children, particularly for poorer women, and this plays asignificant role in the intergenerational transmission of income and socio-economicstatus ( [Currie and Moretti, 2007, Currie, 2009]).

But early life circumstances also refer to events that occurred prior to birth,in the prenatal period. In medicine, the fetal origins hypothesis, put forward byBarker ( [Barker, 1990]), postulates that numerous adult diseases have their ori-

1

gins in the prenatal period, when the fetus adapts to the environmental cues and tothe predicted postnatal environment. This hypothesis has since been explored ineconomics, and it was shown that in utero conditions, and especially adverse fetalshocks, can have significant deleterious effects both in the short run as well as inthe long run on human capital measures that also extend beyond health outcomes.

As such, birth weight has been found to be negatively affected by nutritionaldeprivation during pregnancy caused by, for example, Ramadan observance ( [Al-mond and Mazumder, 2011]) or financial hardship in economic downturns ( [Boz-zoli and Quintana-Domeque, 2014] ). Maternal exposure during pregnancy to psy-chological stress, such as that induced by armed conflict ( [Camacho, 2008]) ordistressing economic news ( [Carlson, 2015]), also lowers birth weight, and similareffects have been shown as a consequence of prenatal exposure to environmentalpollutants ( [Currie and Schmieder, 2009]). However, these deleterious effects atbirth are conditional on the child being carried to term. Significant fetal insultscan have such a large negative effect that the least fit fetuses are miscarried, andthe culling mechanism leads to better average health status in the surviving cohort.This selection in utero phenomenon has been documented in response severelystressful events, such as civil conflicts ( [Valente, 2015]). However, the effects offetal insults may remain latent until adulthood and affect non-health outcomes. Forexample, [Almond et al., 2009] finds that exposure while in utero to the radioactivefallout from Chernobyl led to worse performance in secondary school, but did notaffect heath status. Very importantly, it seems that that early life shocks occur morefrequently and have larger impacts on children from poorer families ( [Case et al.,2002]).

This body of literature that shows that early life environment, as the sum of preand post birth influences, has a large effect on the accumulation of human capitalhas then important policy implications. Policies and remediation programs couldbe designed to reduce the incidence of shocks or adverse effects that impact childhealth, either before or after birth. Reducing inequalities in child health in gen-eral, and in health at birth in particular, could be an effective means of increasingthe equality of opportunity and improving adult outcomes and breaking the inter-generational transmission of poor outcomes. It is thus important to understand ingreater depth the determinants of early child health, and how social and economicpolicies directly or indirectly affect this essential part of the human capital stock.The work in this thesis contributes to this strand of the literature by investigating

2

further the determinants of early child health and the effects on health at birth ofspecific policies and programs, in the Romanian context.

Early life environment is also implicitly related to the fertility decisions of par-ents, and the parental investment response to the previously discusses shocks. Theseminal work of Becker ( [Becker, 1960]) outlines the trade-off between quantityand quality of children and analyzes how the demand for fertility is affected bythe cost of children, and the relationship between income and fertility. Parental in-vestments during early childhood may compensate or reinforce the effects of earlyshocks; [Hsin, 2012] finds using a siblings sample that less educated mothers devotemore total time to heavier birth weight children, whereas better educated mothersdevote more total time to lower birth weight children, with the compensating ef-fects being much larger than the reinforcing effects. The last chapter of this thesisinvestigates the role of financial incentives on fertility, reproductive behavior andearly investments in child health by exploiting a major change in the maternityleave benefits policy in Romania.

1.2 Romanian context

Romania, European Unions seventh largest member state by population size butsecond to last by GDP per capita, provides excellent opportunities to study the de-terminants of early life environment and fertility. After several decades under acommunist regime that enforced drastic pro-natalist measures, which included anabortion ban and penalties for childless couples, the country transitioned to democ-racy and liberalized fertility choices, by re-legalizing abortion and the use of fertil-ity control methods. The regime shift led to a very large drop in the total fertilityrate, from 2.30 in the late 1980s to 1.30 just a decade later, and the highest abor-tion rates in Europe. These spiked in 1990 to a staggering 3152 abortions per 1000live births in 1990 and then steadily decreased to roughly 1000 abortions per 1000births in 2000 and halved again by 2010, but remained double relative to the Euro-pean Union average. Family policies were updated to accommodate the new marketeconomy, and suffered several substantial changes over the last decades. While thescope of benefits for families increased up to 2010, Romania experienced decreas-ing fertility rates and negative population growth.

In terms of child health at birth, although significant progress has been reg-istered, Romania still registers high low birth weight rates and infant death rates

3

relative to the European and US averages. The average low birth weight rate in Ro-mania over the last two decades has been 9%, compared to an EU average of 6.5%.For infant mortality, the rate per 1000 live births decreased steadily from over 20in the early 1990s to little over 10 in 2010, whereas the average rate in the otherEuropean Union countries rate decreased over the same period from roughly 10 to4. These statistics indicate that there is a large scope both for direct and indirectinterventions that target child health at birth and other components of the early liveenvironment.

Despite the increasing checks and balances that are being placed on the politicalsystem, the Government, which holds the executive power, frequently intervenes inthe legislative process by de facto introducing new laws or changing the content ofalready active ones through Emergency Ordinances. As such, over the last decadesthe major laws governing health, education or social assistance changed radicallyand suddenly several times, without parliamentary consensus or prior consultationswith the civil society. These provide excellent sources of quasi-natural experimentsinduced by unexpected policy changes, that are decided upon and implementedover very short periods of time, which most often do not provide the individualsthe opportunity to adjust their behavior prior to the change.

1.3 Summary of the thesis

This thesis contains three empirical papers that explore the socio-economic de-terminants of early child health, using quasi-natural experiments induced by policychanges. The first paper investigates the effects of income shocks during pregnancyon the health at birth of children in utero at the time of the shock and finds evidencethat the selection effects of economic shocks can be larger than the scarring ef-fects. The second paper evaluates the effects of a public health program targetedat a very disadvantaged ethnic minority that provided information to increase thehealth status of women and children; we show that information provision increasesthe take-up of prenatal care, but may be insufficient to improve childrens healthoutcomes at birth. The third paper assesses the effects of financial incentives, inthe form of maternity leave benefits, on fertility behaviors and early investmentsin child heath. We find effects in line with the Becker model of fertility ( [Becker,1991]) and the trade-off between quantity and quality of children.

4

Paper I, "Austerity Measures and Infant Health. Lessons from an UnexpectedWage Cut Policy" (with Andreea Mitrut), analyzes the effects of an exogenous in-come shock during pregnancy on health at birth of children in utero. We use thequasi-experimental setting created by a major (25%) and unexpected wage cut aus-terity measure that affected all public sector employees in Romania in 2010. Weuse all registered births in Romania over the period 2007-2010 in a double differ-ence design, where we use out of the labor force mothers as the control group. Ourmain findings indicate an overall improvement in health at birth, measured as theprobability of low birth weight, for boys exposed to the shock in early gestation,whereas there are no effects on girls of any gestational age. Additionally, we finda decreased sex ratio at birth among early exposed children. These findings areconsistent with the selection in utero theory hypothesizing that maternal exposureto a significant shock early in gestation preponderantly selects against frail male fe-tuses, with healthier survivors being carried to term. This is the first economic studyto find evidence consistent with selection in utero induced by economic shocks.

In Paper II, "Bridging the Gap for Roma Women: The Effects of a Health Medi-ation Program on Roma Prenatal Care and Child Health" ( with Andreea Mitrut),we investigate the effects of a large public health program, targeted at a highlydisadvantaged ethnic group, on maternal and child health. Roma, Europes largestminority, face poverty, social exclusion and life-long inequalities, despite the in-tensified efforts to alleviate their plight. The Roma Health Mediation program wasdesigned as a large-scale public health program, aiming to improve the health statusof pregnant Roma women and their children with the help of trained Roma healthmediators from the local community. Mostly through home visits, the mediatorsprovided information and basic health education, and facilitated the communica-tion between Roma ethnics and medical practitioners. Using unique register datafrom Romania, we exploit the spatial and temporal variation in the implementationdates of the program at the locality level to identify the causal effects of the RomaHealth Mediation program on prenatal care take-up rates and child health at birthfor Roma ethnics. We find that the program had a very large impact on the take-up rates of prenatal care services, which increase with time since implementation.Despite the large improvements in prenatal care take-up rates, we find no changesin the health outcomes at birth of Roma children, in line with previous literature onthe limited effect of prenatal care in non-problematic pregnancies on health at birth.However, we do find evidence of a decreased number of stillbirths and infant deaths

5

at the locality level after the program implementation, but due to data limitations,it is unknown whether these were Roma ethnics.

Paper III, "The Effects of Financial Incentives on Fertility and Early Invest-ments in Child Health" identifies the impact of financial incentives on fertility be-havior and early child outcomes using an unexpected change in the way maternityleave benefits were awarded. The change entailed the switch from proportional(equal to 85% of the mothers pre-birth earnings) to fixed benefits, with the levelof the fixed benefits larger than the wage income of most employed women. Us-ing data from the Romanian Reproductive Health Survey collected one and a halfyears after the policy change announcement, I explore the entire spectrum of in-dividual level decisions related to fertility: decision to conceive, decision to carrythe pregnancy to term, and several important outcomes conditional on live birth(maternal behavior during pregnancy, child health at birth and early investmentsin child health). I employ a double difference identification design in which em-ployed women are the treatment group and out of the labor force women are thecontrol group. Although marginally insignificant, the main findings suggest thatthe substantial increase in the financial incentives led to an increase in concep-tion rates and a decrease in the probability of abortion, especially for women frompoorer households, who benefited more from the policy change. Employed moth-ers who benefited from the change appear to have worse prenatal behaviors buthave children with better health outcomes at birth. Employed mothers who weredisadvantaged by the change make more investments in child health.

Bibliography

[Almond et al., 2009] Almond, D., Edlund, L., and Palme, M. (2009). Chernobyls subclin-

ical legacy: Prenatal exposure to radioactive fallout and school outcomes in Sweden. The

Quarterly Journal of Economics, 124(4):17291772.

[Almond and Mazumder, 2011] Almond, D. and Mazumder, B. (2011). Health capital and

the prenatal environment: the effect of Ramadan observance during pregnancy. American

Economic Journal-Applied Economics, 3(4):56.

[Barker, 1990] Barker, D. J. (1990). The fetal and infant origins of adult disease. BMJ:

British Medical Journal, 301(6761):1111.

[Becker, 1960] Becker, G. S. (1960). An economic analysis of fertility. In Demographic and

economic change in developed countries, pages 209240. Columbia University Press.

6

[Becker, 1991] Becker, G. S. (1991). A treatise on the family. Harvard University Press,

Cambridge, Mass.

[Black et al., 2007] Black, S. E., Devereux, P. J., and Salvanes, K. G. (2007). From the

cradle to the labor market? The effect of birth weight on adult outcomes. The Quarterly

Journal of Economics, 122(1):409439.

[Bozzoli and Quintana-Domeque, 2014] Bozzoli, C. and Quintana-Domeque, C. (2014).

The weight of the crisis: Evidence from newborns in Argentina. Review of Economics

and Statistics, 96(3):550562.

[Camacho, 2008] Camacho, A. (2008). Stress and birth weight: evidence from terrorist

attacks. The American Economic Review, pages 511515.

[Carlson, 2015] Carlson, K. (2015). Fear itself: The effects of distressing economic news

on birth outcomes. Journal of Health Economics, 41(0):117 132.

[Case et al., 2002] Case, A., Lubotsky, D., and Paxson, C. (2002). Economic status

and health in childhood: The origins of the gradient. American Economic Review,

92(5):13081334.

[Currie, 2009] Currie, J. (2009). Healthy, wealthy, and wise: Socioeconomic status, poor

health in childhood, and human capital development. Journal of Economic Literature,

47(1):87122.

[Currie and Almond, 2011] Currie, J. and Almond, D. (2011). Chapter 15 Human capital

development before age five. volume 4, Part B of Handbook of Labor Economics, pages

1315 1486. Elsevier.

[Currie and Hyson, 1999] Currie, J. and Hyson, R. (1999). Is the impact of health shocks

cushioned by socioeconomic status? The case of low birthweight. American Economic

Review, 89(2):245250.

[Currie and Moretti, 2007] Currie, J. and Moretti, E. (2007). Biology as destiny? Short and

long run determinants of intergenerational transmission of birth weight. Journal of Labor

Economics, 25(2):pp. 231264.

[Currie and Schmieder, 2009] Currie, J. and Schmieder, J. F. (2009). Fetal exposures to

toxic releases and infant health. American Economic Review, 99(2):17783.

[Hsin, 2012] Hsin, A. (2012). Is biology destiny? Birth weight and differential parental

treatment. Demography, 49(4):13851405.

[Valente, 2015] Valente, C. (2015). Civil conflict, gender-specific fetal loss, and selection:

A new test of the Trivers-Willard hypothesis. Journal of Health Economics, 39(0):3150.

7

8

Paper I

Chapter 2

Austerity Measures and Infant Health. Lessons from anUnexpected Wage Cut

Simona Bejenariu Andreea Mitrut

AbstractWe investigate the effects on the health at birth of children exposed in utero to a ma-jor (25%) and unexpected wage cut austerity measure that affected all public sectoremployees in Romania in 2010. Our findings suggest an overall improvement inhealth at birth for boys exposed to the shock in early gestation and a decreased sexratio at birth among early exposed children. These findings seem consistent withthe selection in utero theory hypothesizing that maternal exposure to a significantshock early in gestation preponderantly selects against frail male fetuses.

JEL classification codes: I19, J13, J38, I38Keywords: austerity; fetal shock; health at birth; selection in utero; Romania

Department of Economics, University of Gothenburg. simona.bejenariu@economics.gu.seDepartment of Economics, University of Gothenburg. andreea.mitrut@economics.gu.se;

Corresponding author. Mailing address: Department of Economics, University of Gothenburg, Box 640,SE 40530, Sweden. Mitrut gratefully acknowledges support from Jan Wallanders and Tom HedeliusFond. We thank Marcus Eliason, Lennart Flood, Mikael Lindahl, Peter Nilsson, Katarina Nordblom,Henry Ohlsson, Ola Olsson, Olof Johansson-Stenman, Climent Quintana-Domeque, Matthias Sutter,Mns Sderbom, and participants at various workshops and seminars.

11

2.1 Introduction

(How) Are the unborn children affected by austerity measures? While childrenin utero are not the intended target of the recent austerity programs, they may benegatively affected as governments across many countries take action to eliminateunsustainable budget deficits. Within the framework of the fetal origin hypothesisput forward by [Barker, 1990], recent evidence shows that, indeed, disruptions inprenatal conditions, caused by fetal shocks,1 have scarring, life-long consequences(see [Almond and Currie, 2011, Almond and Mazumder, 2011]). While prior workhas found that extreme events can substantially affect fetal health,2 little is stillknown about the effects of shocks induced by economic phenomena. Understand-ing whether and how economic downturns affect fetal development is especiallyrelevant in the aftermath of the Great Recession, which caused significant economicdisruptions and forced governments to impose harsh austerity measures. Public sec-tor wages were frozen in numerous European countries, while others implementedwage cut policies.3 In this paper we exploit the most drastic wage cut austeritymeasure implemented in Europe, entailing a 25% cut in wages and in all the addi-tional benefits for all public sector employees in Romania starting July 1st, 2010.This led to a drop of 60.1 percentage points in the public sector wage premium.4

This unexpected and major wage cut provides an excellent setting to explore theeffects of an exogenous income shock on health outcomes at birth.

The effects of economic phenomena on fetal environment are, in general, quitedifficult to disentangle as their timing is usually diffuse, lacking a precise onsetdate, and they may affect fetal health through multiple channels simultaneously( [Almond and Currie, 2011]). During economic hardship, individuals may reduceexpenditures on consumption goods, and nutritional restrictions may affect the un-born child. At the same time, the countercyclical pattern of consumption of health-

1Fetal shocks are defined broadly as events that alter the fetal environment, and give rise to fetalstressors that may induce developmental adaptations in the unborn child, as they signal a change in thepredicted postnatal environment ( [Gluckman and Hanson, 2005]).

2E.g. civil and military conflicts ( [Catalano, 2003, Mansour and Rees, 2012, Valente, 2015]), naturaldisasters ( [Almond et al., 2007]), terrorist acts ( [Glynn et al., 2001, Camacho, 2008]) and pandemics( [Almond, 2006]).

3Wage cuts were implemented in: Romania (25%, 2010), Czech Republic (10%, 2011), Estonia(6%, 2009-2010), Greece (20%, 2012), Ireland (5%, 2010), Hungary (7%, 2008-2010), Latvia (15%,2009-2010), Lithuania (15%, 2009-2010), Portugal (5%, 2011), Slovenia (4%, 2011), Spain (5%, 2010).We discuss in Section 2 that the wage cut austerity policy was most likely not anticipated in Romania.

4The public sector wage premium fell from +44.5% in 2009 to -15.6% in 2010 (a loss of 60.1percentage points) (source: Industrial Relations in Europe 2012 Report, European Commission).

12

damaging goods and the decrease of the opportunity cost of health-improving be-haviour may offset the negative effects and lead to better infant heath at birth. Inaddition, maternal prenatal stress, caused by the financial insecurity entailed byeconomic shocks, may have either scarring and/or culling effects, leading to anambiguous net effect of economic shocks on health at birth, depending on a widearray of factors. Thus, some studies find evidence of deteriorating health outcomesat birth ( [Bozzoli and Quintana-Domeque, 2013, Paxson and Schady, 2005, Bur-lando, 2010, Lindo, 2011]), whereas others find that the effects of improvementsin risk-related behavior during pregnancy and maternal selection prevail over thescarring effects, the net result being an improvement of the health of in utero ex-posed children ( [Dehejia and Lleras-Muney, 2004]). Additionally, the sex-ratio atbirth has also been found to respond to economic circumstances ( [Catalano andBruckner, 2005, Catalano, 2003, Catalano et al., 2009]). [Bozzoli and Quintana-Domeque, 2013] document the pro-cyclical effects of economic fluctuations in Ar-gentina on the birth outcomes of children, noting that birth weights are sensitive tomacroeconomic fluctuations during the third trimester of pregnancy via the nutri-tional deprivations channel and during the first trimester of pregnancy via the ma-ternal stress channel. [Almond et al., 2011] look at the effect of the Food StampsProgram in the US as a positive shock in utero and find improvements in healthoutcomes at birth.

However, these effects observed at birth and/or later on in life are, in fact, con-ditioned on the fetus surviving the pre-birth period. Medical literature finds thatsignificant prenatal maternal stress, especially during early gestation, may inducea selective mortality of the least fit fetuses through increased miscarriages. Thisprocess, known as selection in utero, may yield a positive selection of those thatare carried to term, visible in an improvement in the health outcomes of the af-fected cohort, with weak male fetuses significantly more affected than female fe-tuses ( [Hobel et al., 1999, Catalano et al., 2009]).

This paper contributes to the literature on the impacts of (negative) economicshocks on the health outcomes at birth by exploring a unique austerity measure,unexpected in its magnitude (25% cut in wages and in all benefits) and timing (tostart with July 1st, 2010, after being firstly announced on May 7th, 2010). Thedistinct occurrence of the shock eliminates the problem posed by diffuse timing orendogenous income reductions and allows us to pursue a clean identification strat-egy to infer the causal effects of an income shock on birth outcomes of children

13

exposed in utero. Our findings indicate that maternal exposure to economic inse-curity and income loss may lead to what appears to be selection in utero. Whilesome evidence of selection in utero induced by economic fluctuations is providedin epidemiology and demography (e.g., [Catalano et al., 2010], [Catalano, 2003]),this is the first study to show consistent evidence that unexpected austerity mea-sures (entailing large income cuts) experienced during early pregnancy may lead tofewer, but apparently healthier boys at birth.5

Our main empirical strategy to assess the impact of the unexpected incomeshock on health outcomes at birth is a difference-in-difference (DD) specification.We use the Romanian Vital Statistics Natality files containing the universe of birthsfor the period 2007-2010 and compare outcomes at birth for children in utero at thetime of the policy announcement belongings to mothers employed in the public sec-tor and housewife (or alternatively, privately employed) mothers in 2010, relative toearlier years. We will focus on women already pregnant at the time of the austerityannouncement to mitigate the concern related to the change in the composition offamilies choosing to conceive. Unfortunately, we are not able to clearly disentanglebetween the impact of announcement per se and the wage cut two months after.

Our main findings suggest an overall improvement in health at birth as mea-sured by a 1.4 percentage point (pp) decrease (13% of the mean) in the probabilityof low birth weight of children exposed to the shock during their 1st trimester ofgestation. We find significant improvements in health at birth exclusively for boysand not for girls, driven by significant effects for those who have been exposed tothe shock starting with very early developmental stages (1st trimester), a decreaseof 2.9 pp (29% of the mean) in the probability of low birth weight. This effectis particularly large for boys belonging to highly educated mothers. We also findindications of a decreased sex-ratio at birth of about 3.3 percentage points (6.7% ofthe mean) for the same sub-sample of children. Our results hold to a wide series offalsification and robustness tests, including a mothers fixed effects specification.

Using complimentary datasets, we investigate the potential mechanisms throughwhich the austerity measures affected health at birth and find evidence which seemto indicate that selection in utero due to maternal prenatal exposure to the policyshock resulted in a healthier but smaller cohort of boys. Yet, we cannot fully un-

5Within economics, [Valente, 2015] documents selection in utero following a civil conflict in Nepal,while [Nilsson, 2014] finds that boys exposed early in utero to an increase in the availability of alcoholin Sweden are more likely to be spontaneously aborted.

14

derstand whether the fetal stressors are related directly to stress per se (through in-crease in cortisol level) and/or indirectly, through higher intake of alcohol or smok-ing consumption. Overall, our findings are consistent with the medical literaturethat has established that weaker males are more vulnerable to adverse conditionsin utero and that maternal prenatal stressors raises the fitness criterion of childrenin utero. The remainder of the paper is organised as follows: Section 2.2 depictsthe Romanian context in which the policy change occurred, and presents the datawe are using. Section 2.3 describes the empirical strategy, and presents the mainresults, followed by several sensitivity checks. In Section 2.4 we discuss the po-tential mechanisms through which an income shock may affect birth outcomes andfurther test these mechanisms in Section 2.5. Section 2.6 contains a series of furtherrobustness checks that support our main results. Section 2.7 concludes.

2.2 Background and data

2.2.1 The Romanian context

Romania experienced sizable economic and politic insecurity throughout most ofits post-communist period.6 Thus, the international financial crisis that unfoldedin the autumn of 2008 was taken lightly in Romania: politicians invoked a decou-pling of the Romanian economy from the world markets, and the public opinionwas moderate in its expectations: the autumn 2008 Euro-barometer showed thatmore than half of respondents anticipated no change or even an improvement inthe general economic situation of Romania, with the same attitude prevailing in the2009 waves of the survey.7

The first political signs of the recognition of the deteriorating state of the Ro-manian economy came in March 2009, when the Government initiated discussionswith the IMF. After signing a stand-by accord in June 2009, politicians promotedthe agreement as an opportunity for state reorganization, but subsequent proposedmeasures were mild and noncontroversial. Moreover, the political class transmittedan overall confident message in the lead-up period to the presidential elections of

6Although negative growth rates were replaced by high growth rates beginning in 1999, they wereaccompanied by high inflation rates and significant public deficit. In 2000, when the GDP growth rateturned positive, the annual inflation rate was over 40%, whereas in 2004, when the GDP annual growthrate reached a peak of almost 9%, the annual inflation rate was still above 10%.

7http://ec.europa.eu/public_opinion/cf/

15

http://ec.europa.eu/public_opinion/cf/

December 2009. After being re-elected, the incumbent President declared that "(...)we expect significant growth in the first part of 2010".8

In this context, the Presidents announcement on the national TV, on May 7th,2010, that public sector wages and social security benefits would be cut was unex-pected and gave rise to widespread social unrest and political dispute. The decisionwas made by the Government and the President after the latest round of negotia-tions with the IMF and was not preceded by any discussions in the Parliament orwith social partners, nor was publicly mentioned as a potential policy. The mea-sures, involving a 25% cut in wages for all public sector employees, the revocationof most of their financial and in-kind incentives and a 15% cut in unemployment,maternity leave benefits and several other social security benefits, were aimed atre-establishing the budgetary balance agreed to with the IMF. Thus, for pregnantwomen employed in the public sector at the time of the announcement (our treat-ment group), the austerity policy had a threefold effect: a monthly income drop dueto the wage and benefits cut; a decrease in the annual average wage income whichwould lead to a lower (forthcoming) child care allowance, calculated as 85% of theaverage income obtained over the 12 calendar months preceding the birth of thechild; and a 15% cut in the recalculated child care allowance to be received afterbirth.

One month after the announcement of the austerity measures, the Finance Min-ister gave a speech pertaining to the delusional nature of the governments previousstatements on the economic status of the country and on the completely unexpectednature of the policy: As a Finance Minister I am telling you that we could havelied six more months, we could have borrowed for six months, [...] and could havewaited six months to see what happens. The fact that what we are doing entailsa political risk that nobody imagined a month and a half ago shows a completeresponsibility of this Government towards the Romanian citizens.9 He was dis-missed shortly after.

The measures were included in a set of legislative projects drafted by the Gov-ernment soon after the Presidents announcement and forwarded to the Parliamentto be adopted through a special procedure that circumvented the regular and lengthy

8http://goo.gl/sMcVEV (in Romanian). Early in 2010, the Government adopted a graver atti-tude toward the worsening economic crisis as the IMF required concrete actions to reduce the significantbudget deficit. As such, on March 16th, 2010, the Prime Minister presented in front of the Parliamentthe anti-crisis measures that were being implemented, all as economic stimulus, aimed at improving thebusiness environment and reducing tax evasion.

9http://goo.gl/bJNNYr (in Romanian)

16

http://goo.gl/sMcVEVhttp://goo.gl/bJNNYr

law making procedures.10 On June 30th, the President promulgated the laws, whichcame in effect July 1st, with an initial duration of 6 months, but in fact in January2011 public sector wages were not restored to their initial level.11

Overall, it is safe to assume that the austerity measures were not anticipated,in both their unprecedented scope and magnitude, or their timing. In our empiricalstrategy we will focus on women working in the public sector, already pregnantat the time of the austerity announcement, to mitigate the concern related to thechange in the composition of families choosing to conceive. Even though the aus-terity measures were unanticipated, we cannot exclude written on the wall.12 Thepossible selections into fertility will be addressed later in the paper.

The public sector wage cuts affect females significantly more than men due tothe structure of the public sector employment.13. In Romania, the publicly em-ployed women are concentrated in Health, Social Services and Education sectors,and had, even before the austerity measure, lower average wages both relative tothe private sector and to other public, male dominated sectors such as Local Ad-ministration and Defense.14 In addition, recent evidence shows that the insecu-rity coupled with the economic crisis has worsened the perception of work-relatedstress in all European countries in general, and in Romania, already ranked high,in particular, making the publicly employed women the most affected by the wagecut, both in monetary and psychological distress terms (see [Vrga et al., 2012]).

10The Romanian Constitution allows, as an exception, that the Government assumes responsibilityfor a specific law in front of the Parliament, with the law under consideration being adopted by defaultif the Government is not dismissed in the first 3 days by means of an adopted censorship motion. TheParliament can withdraw the trust awarded to the Government by adopting a censorship motion, whichnecessarily means that the Government is dissolved, the law proposed is not adopted and a new Gov-ernment needs to be invested. After the Government assumed responsibility on the Austerity Laws, acensorship motion was initiated by the opposition parties in the Parliament but because of a tight major-ity of the governing coalition, the censorship motion was not adopted (though by a very close margin)and the Laws were passed in a slightly modified version.

11It is important to distinguish between a permanent and a temporary wage cut: transitory changes inwages have no effect on lifetime income or on total fertility though they may affect the timing of fertility,while a permanent wage cut has an ambiguous effect as it may decrease the relative cost of childrenwhich, in turn, may increase the demand for children or, because of a lower income, it may decreasethe demand for children; [Becker, 1965, Heckman and Walker, 1990]. Even if temporary, householdsmight respond as though these changes are permanent if people are myopic or uncertain about the natureof the changes ( [Dehejia and Lleras-Muney, 2004]). This was most likely the case in Romania, withmost households perceiving the wage cut as permanent, because of numerous inconsistent enforcementof laws.

12At that time Romania experienced an increase in the unemployment rates in the private sector, thatrose from a relatively stable level of 4 to 5% before 2009 to 8% in March 2010.

13Source: Industrial Relations in Europe 2010 Report, European Commission14Source: Statistics Romania.

17

2.2.2 The impact of the austerity measures at household level

To understand the size of the impact of the austerity, we use the Romanian House-hold Budget Survey (RHBS), the main tool of assessing population expendituresand revenues, covering about 30,000 households/year and containing detailed in-come and expenditure information. We compare here households with at least onepublicly employed member and households with no publicly employed members,just before (January-July 2010) and after (August-December 2010) the austeritymeasures implementation. The results in Table 2.1 indicate a significant decreasein household wage related income of 16.7% and in total household income of about7% for households with at least one publicly employed member.15 Not surprisingly,the wage related income drop is larger for high-educated households (about 21.7%in column 2) because the high-educated publicly employed members were morelikely to attract more wage related income (through bonuses, in-kinds wage re-lated transfers) which were also annulled. Overall, the households affected by theshock seem to have no significant changes in food-related (column 4) or alcoholand cigarettes (column 5) expenditures, but they have significantly reduce non-food (column 6) and services expenditures (column 7). Finally, column (8) seem toindicate that households react to the wage shock by decreasing the (formal) savingswith about 11.9%.

2.2.3 Data and working sample

In our main empirical exercise we use the Vital Statistics Natality (VSN) recordsfor years 2007 through to 2010,16 as our main dataset. The VSN records coverthe universe of live births, with detailed information about the newborn and thesocio-economic characteristics of the parents, recorded at the time of the birth:(a) characteristics of the child: date of birth, gender, ethnicity, whether singletonor multiple birth, birth weight and duration of gestation in number of weeks; (b)characteristics of the mother: date of birth, occupational status, education, marital

15It is not surprising that the wage drop was not 25% (or higher) as the data provides information atthe household level. Also, we show these results only for urban households (see the explanations in thenext section).

16In 2011, Statistics Romania changed the data registration process for the VSN, and no longer col-lects information on a wide array of maternal and child characteristics which we use in the currentanalysis. Also, we cannot use pre-2007 VSN datasets since we do not have access to RHBS datasetsprior to 2007, hence we cannot calculate the predicted probability of public employment, as will bediscussed shortly.

18

status, county and locality of residence, and mothers fertility history: total numberof births, number of children born alive, fetal deaths, month of first prenatal check-up and an indicator for home delivery; (c) characteristics of the father: date of birthand his occupational status.

We restrict our sample to mothers between 16 and 45 years of age that werepregnant on May 7th each year and exclude multiple births, which leaves us witha sample of 846,778 births over the period 2007-2010. In the baseline estimationswe will focus on children born from mothers living in urban areas, accounting for465,754 of all births. Given the nature of the policy change, there are reasons to ex-pect that effects would be concentrated among urban rather than rural households.Firstly, among the employed women of fertile age, living in rural areas, only about8% work in the public sector compared to about 30% of the employed women fromurban areas (source: RHBS). Secondly, we suspect that the wage cut policy affectedthe rural households much less relative to the urban households because in wageincome represents less than 20% of the total household income in rural areas, com-pared to an average of 60% for families living in urban areas ( [Firici and Thomson,2002]).17 Even though our empirical analysis discusses urban households, we alsoshow that our main results hold when we look at all households. Summary statis-tics for our main variables for the urban mothers are found in Table 2.2, columnblock 1.18

A key variable in our empirical specification is the mothers occupational status.The VSN records the mothers occupational status using the following categories:employed, entrepreneur, self-employed in agricultural activities, self-employed innon-agricultural activities, unemployed, housewife, retiree, and other situations.However, the employed category does not differentiate between public and privatesector of employment.

Because the policy specifically targeted the public sector employees, we pro-ceed by making use of the RHBS for the 2007-2010 period, which includes thesame socio-economic characteristics as the VSN and in addition records the sectorof employment. We estimate a conditional probability that an employed womanworks in the public (vs. the private) sector and conduct out of sample estimationto assign mothers in the VSN probabilities of public employment (we will come

17Agricultural own-production income is estimated as high as 46% for rural households and about13% for urban households ( [Firici and Thomson, 2002]).

18Appendix Table 2.12 in the Appendix A shows the descriptive statistics for the urban and ruralsample.

19

back to this issue in Section 3.2.2. and in Appendix B). Next, to define our treat-ment group, we make use of information provided by the Romanian Ministry ofLabour, Family and Social Protection (MLFSP) regarding the recipients of childcare allowance.19 At the end of 2010, among the employed mothers receiving childcare allowance, 20% were working in the public sector and 80% in the private sec-tor.20 We use this percentile split and classify as publicly employed the employedmothers with the 20% highest predicted probabilities. We will conduct several sen-sitivity analyses with respect to the choice of the threshold percentile (includingestimating a model using the continuous measure of the probability rather than abinary indicator) and the assignment into the treatment group (see Section 3.2.2and Appendix B).

The main characteristics of the publicly employed mothers as defined by the20-80 split are shown in Table 2.2, column block 2. Compared to the sample of allemployed mothers, shown in column block 3, the publicly employed mothers are,on average, older, more likely to be married and more educated. Reassuringly, thiscomposition matches very well that of the publicly employed women in the RHBSdata.21 Relative to all mothers or to all employed mothers, the publicly employedmothers seem to have healthier children as measured by birth weight and gestationlength. In column block 4 we show the main characteristics of the housewivesmothers, accounting for about 30% of all mothers in urban area.22 Housewivesmothers are, on average, younger, less likely to be married, lower educated andhave children with worse outcomes at birth relative to mothers in column blocks 1to 3.

At this stage we also check possible anticipatory effects in terms of selectioninto motherhood of the austerity measures. Overall, from Table 2.2 we observe that

19Child care allowance is awarded to either one of the parents who has obtained any form of taxableincome in the 12 months preceding the birth of the child. Basically all employed mothers receive thisallowance.

20MLFSP does not hold centralized information on the number of recipients of child care allowanceby the childs month, year and county of birth and mothers sector of employment.

21Albeit a small sample, among the 230 mothers (with a child one year old or less between 2007 and2010) employed in the public sector, 77% have high education, while only 6% have secondary education.Among the employed women in the private sector who have recently become mothers (1,102), only30% have higher education, 40% have high-school education and 22% have secondary education. Thismatches very well with the composition we obtain in our treatment group based on the 80-20 split.

22For the entire sample including urban and rural women, the occupational structure reveals that47.8% of women giving birth in 2010 are employed; 42.6%, housewives; 0.15%, business-owners; 1%,self-employed in non-agricultural activities; 0.2%, self-employed in agriculture; 1.8%, unemployed;0.2%, pensioners; and 6.25%, other situations. This structure is quite stable over the years and the areaof residence.

20

employed mothers who give birth in later years in urban areas seem to be bettereducated (more likely to have higher education) which may be due to a positiveselection into motherhood, but also because a well-recognised trend in educationin Romania.23 The publicly employed mothers, even though are on average moreeducated compared to the other occupational categories, in 2010 (relative to before)they are less likely to have a higher degree and more likely to only have a post-highschool degree, suggesting a negative selection.

To address the issue more formally, in Table 2.3, for each occupational categorywe run regressions with mothers observable characteristics as outcomes. Overall,mothers pregnant on May 7th, 2010, relative to those pregnant before, are morelikely to be more educated and slightly older. This is also true for the housewivesand particularly for the privately employed mothers. The effects are significant andquite large as a percentage change from the mean. However, publicly employedwomen pregnant at the time of the announcement seem to be less educated (morehave only secondary or high school and fewer have a higher education) and they areless likely to be married. Albeit statistically significant, the changes relative to themean are not as large as for the privately employed or for the housewives mothers.24

Overall, our results indicate that, even though the austerity measures were mostlikely unanticipated, the overall economic context has influenced the fertility timingdecision of Romanian women and has altered the composition of mothers becomingpregnant. These findings are in line with other studies (see [Dehejia and Lleras-Muney, 2004] for the US) that show that in turbulent economic times, we mayobserve an increase fertility of low-skilled women (as measured by education) anda negative selection for the high-skilled ones.25 It is important to note that using our

23See Appendix A, Figure 2.4. The significant increase in the number of higher educated individualsis due to the massive increase in the number of private universities. Figure 2.5 shows that over the 2003-2010 period, while the proportion of employed mothers with primary education is relatively constantacross years, there is an increase in the employed mothers with higher education matching the decreaseof the employed mothers with secondary education.

24An alternative way to analyse the selection into fertility issue is to estimate the baseline differencein difference regressions comparing the characteristics of the publicly employed mothers with those ofthe housewife mothers, pregnant at the time of the austerity measures announcement relative to the sameperiod in previous years. In accordance with the previous findings, we find that relative to housewifemothers, publicly employed mothers from urban areas are less educated (lower probability to havehigher degree and higher probability to have secondary education), younger, less likely to be marriedand have an unemployed husband. The results are presented in Appendix Table 2.13.

25The net effect of an economic shock is theoretically ambiguous and hinges upon the mothers skilldepreciation rate and on whether capital markets are perfect ( [Dehejia and Lleras-Muney, 2004]). Onemay hypothesize that low-skilled women are less likely to have a human capital that depreciates duringa temporary absence from a job during pregnancy and after birth (and assuming that capital markets are

21

difference-in-difference identification strategy, a negative selection in the treatmentgroup and a small positive selection in the control group would bias the resultstowards zero and thus any significant results would not be driven by this selection.

2.3 Identification and main results

2.3.1 Identification strategy

To test whether the austerity measures changed the outcomes at birth of the chil-dren in utero at the time of the announcement (May 7th, 2010) relative to childrenconceived in earlier years, we rely on a difference-in-difference (DD) specification.Our treatment group consists of pregnant women classified as working in the publicsector while our control group consists of pregnant housewives. Thus, we compareoutcomes at birth between children in utero on May 7th, 2010, and May 7th of theprevious years (2007-2009), with mothers working in the public sector and house-wives. Housewife mothers is our preferred control group as they are least likelyto have been affected by the austerity measures: they are out of the labour forceand they do not receive any social assistance benefits (such as unemployment ormaternity leave benefits).26 Moreover, they are the second most numerous groupby mothers occupational status, after employed mothers. We acknowledge thathousewives may not be an ideal control group and therefore we will also considerthe mothers classified as privately employed as an alternative control group. Theyare not our preferred control group because they are also defined based on the 20-80split; moreover, we have also shown in the previous section a substantial (positive)change in the composition of privately employed women who become pregnant in2010 which, most likely, will bias our results towards zero.

We measure health at birth using the low birth weight indicator, defined as abirth weight less than 2,500 grams.27 Our baseline specification, estimated using

perfect); if so, then in low-wage periods, we may observe an increase fertility of low-skilled women.26Housewife is defined as a person engaged in domestic work such as preparing food, maintenance

and home care, domestic industry activities not intended for sale, care and education of children andwho does not receive a formal income.

27Using birth weight as a continuous outcome provides fairly similar results. We focus on low birthweight since it is a more accurate measure of neonatal health and a better predictor for infant health,being the leading cause of neonatal and infant mortality ( [Stein et al., 2006]).

22

ordinary least squares, is the following:

Outcomeimrt = + 1Publici + 2Publici Utero2010i+

t + 1Xi + r + rt + m + crt + imrt (2.1)

where i indexes a child born in month m by a mother living in county r in yeart; Publici is an indicator that equals 1 if the mother of child i works in the publicsector and 0 if she is a housewife (or works in the private sector in an alternativespecification). Our key coefficient is 2, on the interaction between Public andan indicator whether the child was in utero in May 7th 2010. This measures thechange in outcomes after the 2010 announcement relative to earlier years, amongwomen that work in the public sector relative to housewives. t are year indicatorsthat equals 1 if child i was in utero on May 7th in year t; Xi is a vector of con-trol variables for maternal and child characteristics: childs gender, mothers ageat birth and its square, mothers education, ethnicity, marital status, childs parity,number of children alive, indicator for prenatal control, gestation month of the firstprenatal care visit in the current pregnancy and an indicator for home delivery. Ourmain specifications also include the fathers age and its square together with indi-cators for his employment status (whether employed, entrepreneur, self-employedin agricultural activities, self-employed in non-agricultural activities, unemployed,retiree or other situations) at the time of the child birth.28 r are 42 county indica-tors, while rt are linear county specific trends; m are months of birth indicators;with crt, we control for the female unemployment rate in the month of concep-tion for each county and year of conception.29 We cluster the standard errors at thecounty level (42 clusters), even though we get very similar standard errors withoutclustering.

The key identification assumption in a DD framework is that, absent the policychange, we would not observe any difference in our outcomes between publiclyemployed mothers and housewives in 2010 relative to earlier years (the parallel

28Information for the fathers is available regardless of the mothers marital status. However, it ismissing for about 23 percent of the unmarried mothers. For this sample, albeit very small, we haveimputed the missing information with the relevant locality average. Our results are not sensitive toincluding or not this sample.

29The VSN does not include information on mothers drinking or smoking habits. Including controlsfor the average expenditures on cigarettes and alcohol, at the county level, for each year and gestationalmonth c from conception to birth does not change our results. Same if we include the average con-sumption expenditures on food at the county level for each gestational month from conception to birth.Results available.

23

trend assumption). To examine the plausibility of this assumption we add twointeraction terms to the baseline model: the Public indicator interacted with yearindicators Utero2008 and Utero2009.30

Because the literature suggests that the effects of in utero shocks may vary ac-cording to the stages of gestation, we will explore the fact that at the time of theshock children were in different gestational stages. The VSN data contains the ges-tational age in number of weeks at birth and we are able to infer the gestationalage at the date of the austerity announcement.31 Using this information, we splitour sample into the following categories according to their gestational age at May7th, the time of the policy announcement: 1) children in the 1st trimester (up to 12weeks), who were exposed the longest to the policy: to the announcement shock inearly pregnancy and to diminished income later in gestation;32 (2) children in the2nd trimester (13-24 weeks), who were unaffected during the 1st trimester, but ex-posed to policy shock during their 2nd trimester, and to both stress and diminishedincome in late gestation; (3) children in the 3rd trimester (more than 25 weeks),exposed only to the announcement shock in late gestation. It is important to clarifythat in our experiment the de-facto wage cut occurred in early August 2010, whenpublic employees received the wages for July 2010. Hence in the first three monthsfollowing the announcement in utero children were not exposed to reduced incomebut possibly to stress related factors. Due to insufficient variation of policy expo-sure by gestational age, we are not able to clearly disentangle between the effect ofthe austerity announcement per se and that of reduced income. We further discussthis issue in the next sections. Finally, because medical research established thateffects of in utero conditions may depend on the gender of the fetus, we will alsoshow our results separately for boys and girls.

30A graphical illustration of the trends in the outcome of interest is presented in Appendix A, Fig-ure 2.6.

31Having the gestational age in weeks at the time of the announcement allows us to circumvent theproblem of comparing children born in the same month but who were in different developmental stagesat the time of the announcement due to different lengths of gestation.

32Because we cannot use the 2011 VSN, our 1st trimester sample includes only children born in2010. However, for comparability, we do the same restriction for all years and hence this sample isartificially smaller for all years we use.

24

2.3.2 Results

Main estimates

This section presents the baseline results from Equation 2.1 for the low birth weightindicator. Table 2.4 shows the results for the urban households from the DD estima-tion for all (Panel A) and separately for boys (Panel B) and girls (Panel C).33 Eachthree columns of each panel shows the results for children who were in their 1st

trimester, 2nd and 3rd trimester of gestation at the moment of the austerity shock.For each trimester, we first show the interaction term Public*Utero2010 controllingonly for year and county indicators, and county specific trends;34 next we add theindividual level controls; finally, we show the estimated coefficients from the fullyinteracted model, conditional on pre-treatment dynamics.

Panel A shows that the austerity measures affected only children in their 1st

trimester of gestation. The impact of the shock in columns (1)-(2) is negative andsignificant suggesting an reduction of the low birth weight incidence by 2 pp, hencean improved average health. The magnitude becomes 1.4 pp in column (3), after wecontrol for pre-treatment dynamics (13% of the mean). This may be surprising asthese children were exposed to the shock in utero the longest, starting with the veryearly developmental stages. The estimates for the 2009 and 2008 year-specific pub-lic indicators are positive and not statistically significant suggesting that childrenborn from the publicly employed and housewives mothers do not differ significantlyin their evolution of the low birth weight outcome during the pre-treatment years,thus supporting the parallel trend assumption. Moreover, since we employ the sameprocedure to classify publicly employed mothers in all years, the significant coef-ficient for the 2010 interaction cannot be a mechanical result of our imputationmethod. Our results for children in the 2nd and 3rd trimesters of gestation show asimilar pattern, but the magnitude of our main coefficient of interest is smaller andnever significant.35

The results in Panel B indicate a significant decrease of the probability lowbirth weight for the sample of boys in utero in the 1st trimester on May 7th, 2010;

33Appendix Table 2.14 shows the results we also include rural households. The results are in linewith the urban sample, slightly lower in magnitude and significant at a lower level.

34Our results are not sensitive to excluding the county specific trends.35The reason why the Public dummy is insignificant is that a very large share of the publicly employed

mothers have tertiary education, and so the Public dummy will actually capture the high educationdummy. If we exclude the tertiary education among the controls the Public indicator becomes significantand the interaction Public*Utero2010 does not change sign, magnitude or significance.

25

this effect is stable across specifications, of 3.2 pp in columns (1)-(2) and 2.9 ppin column (3) (29% of the mean). This effect holds even though we have shownin the previous section a negative selection among publicly employed mothers in2010 (relative to before and also to the other occupational categories), which wouldrender our results as lower bounds effects of the policy. The 2nd and 3rd trimestersof gestation indicate qualitatively similar results but smaller in magnitude and notsignificant. Finally, the results for girls in Panel C show no effect on low birthweight.

To gain a better understanding about the effects at different gestational ages atthe time of the shock, we use a moving window approach in which we glide thetreatment over cohorts defined in 12 weeks periods, instead of trimesters, at May7th. Figure 2.1 presents the estimated coefficient of interest for each of the 12weeks intervals, for all, boys and girls, respectively, together with the correspond-ing standard errors. For the sample of boys, the effects are decreasing in absolutevalue and remain significant up until the cohort who was 11 to 23 weeks at May 7th,which indicates that children in early second trimester were also affected. For girls,the only significant impacts, in the same direction as for the boys, are observed forgirls who were between 14-26 up to 17-29 weeks. Overall, boys appear signifi-cantly more affected, both in intensity and in number of children affected, with theresults indicating a significant decrease of the probability of low birth weight.

Sensitivity analysis

Before we discuss the possible mechanisms in place, we subject our results to somerobustness tests that address three potential issues: corrected standard errors due tothe generated regressors; the definition of the treatment group; and the compositionof the control group.

a) Corrected standard errors. To account for the fact that we define ourtreatment group based on a generated regressor (i.e. the predicted probability ofpublic employment), we use bootstrapping to estimate the standard errors of theparameters of interest, under the assumption that the OLS estiamtor is consistent(details of how we conducted the bootstrapping procedure are presented in Ap-pendix B4).36 Table 2.5 presents the coefficient of interest, Public*Utero2010, for

36 [Murphy and Topel, 2002]

26

the richest specification presented in Table 2.4 column (3), the robust standard er-rors from our main regression and the bootstrapped standard errors obtained from500 replications. The bootstrapped standard errors for the coefficient of interestare very close to the robust standard errors that we use in our main specification,leading to the same t statistics and the same significance levels for our estimates,leaving the inference unaffected.

b) Sensitivity to the definition of the treatment group. So far, given thelimited information provided by the Romanian MLFSP, we have used the 20-80percentile split of the probabilities of a mothers employment to define our treat-ment. To check the sensitivity of the effects on the low birth weight indicator withrespect to this split, we use different definitions of the treatment group based onvarying the threshold percentile from the 80th to the 50th (i.e., employed moth-ers with predicted probabilities above the threshold percentile are included in thetreatment group). Figure 2.2 confirms that our results, especially for the boys inthe 1st trimester of gestation at the time of the shock, are not sensitive to differ-ent thresholds though and remain negative and significant at 5%, but increasinglybiased towards 0 as we increasingly misclassify the treatment group and includemore privately employed mothers. We also used the predicted probability from theRHBS as a continuous variable and look at the sample of all employed mothers.Our findings are qualitatively similar and indicate that the mothers with higher pre-dicted probability are less likely to have low birth weight boys, but this effect is notstatistically significant. Finally, in Appendix B2 we show some further robustnesschecks.

c) The composition of the control group. As mentioned before, one pos-sible concern is that housewives mothers are not an ideal control group to the em-ployed mothers. We address this issue in several ways.

First, because publicly employed mothers have a high educational level, andthat recent evidence seems to indicate that economic shocks on pregnant womenmay have a different impact according to the mothers SES ( [Bozzoli and Quintana-Domeque, 2013]), we compare only mothers (public and housewives) with higheducation (high school and above). Our results, presented in Table 2.6, show thatthe improvement of the low birth weight indicator we uncovered earlier is driven

27

by the boys belonging to highly educated mothers.37 However, we cannot do thesame comparison for low educated mothers because of a low share of low educatedmothers in the treated group (see Table 2.2).

Secondly, we use as an alternative control group the privately employed moth-ers defined as mothers with the predicted probabilities below the 80th percentile,while keeping the same definition as in the main specification for the publicly em-ployed mothers. Reassuringly, the results in Table 2.7 have a similar pattern asour main outcomes in Table 2.4, especially for the children in the 1st trimester atthe time of the shock, but they are smaller magnitude given the (large) positiveselection into fertility in the private sector.

2.4 Potential mechanisms

In this section we attempt to explain our seemingly counterintuitive results by in-vestigating the potential mechanisms in place. There are three main mechanismsthrough which an income shock generated by an unexpected cut in a pregnantwomans wage may affect childrens outcomes at birth: (1) selection into moth-erhood, (2) nutrition and prenatal care, and (3) prenatal maternal stress.

2.4.1 Selection into fertility and abortions

We try to mitigate some concerns related to changes in the composition of pregnantpublicly employed women by using the fact that the Romanian austerity measureswere unexpected, and by looking at the sample of already pregnant mothers at thetime of the announcement. We have shown in Section 2.2 that some selection intofertility occurred prior to the announcement because of the overall economic situa-tion but, given the nature of the selection, the size and direction of these selectionsdo not invalidate our main results.

Yet, already pregnant women may react to the austerity measures by terminat-ing their pregnancy using abortion. Abortion in Romania is available up to 12 ges-tational weeks. Although we do not have individual data on abortion procedures,

37We have also used a simple matching strategy (nearest neighborhood and 1-to-1 matching, no re-placement) based on pre-treatment characteristics. The effect on the low birth weight indicator is quitesimilar to our baseline estimates, even though less precisely estimated. Additionally, we used a matcheddouble difference specification in which the housewives constituting the control group were weightedwith the inverse of their propensity score; again, the effect on the low birth weight indicator had a similarpattern as our main specification. All results available upon request.

28

we investigate whether the quarterly aggregate number of abortions increased sig-nificantly after the wage cut announcement.38 Reassuringly, we find no significantincrease in the total number of abortions, but we must acknowledge that the abor-tion data is not available by womens employment status.

Because our main findings concern only boys, one may worry that sex selectiveabortion could potentially alter our results. While we are not aware of any evidenceon gender preferences in Romania, one way to formally address this concern is toexamine the pattern of sex-ratio for different child parities over time. In cultureswith sex preferences, sex-ratios are usually normal at first parity but may changewith parity ( [Almond et al., 2009]). Using the VSN data we find no indication ofsex-selection across years or across occupational categories. Finally, our results onlow birth weight hold for a parity larger than two. Moreover, in Romania the childsgender is not routinely detected before 18 gestational weeks whereas abortion ispermitted until the 12th week of gestation, which makes gender-based selectiveabortion, in most cases, impossible.

2.4.2 Nutrition and prenatal care

Prenatal nutrition. A reduced disposable income after July 2010 may lowerthe quantity or the quality of food intake of the mother which, in turn, may leadto an insufficient nutritional supply to the fetus. Such nutritional restrictions mayadversely affect the fetal development, and are often reflected in a higher incidenceof low birth weight, preterm delivery and perinatal morbidity ( [Gluckman andHanson, 2005]; [Abrams et al., 2000], [Fowles, 2004]).39 Importantly, insufficientcaloric intake seems to result in a lower birth weight only in late pregnancy, duringthe 3rd trimester ( [Stephenson and Symonds, 2002]); boys seem, on average, morevulnerable to food shortages than girls ( [Eriksson et al., 2010]). [Almond et al.,2011] show that, in the US, pregnancies exposed to the Food Stamp Program threemonths before birth resulted in an increased birth weight. [Bozzoli and Quintana-Domeque, 2013] find worsening health outcomes at birth for children exposed in

38We use data from the Ministry of Health and estimate a panel fixed effects model in which ourdependent variable is county-by-quarter number of abortions and control for county time trends, sea-sonality and a dummy indicating post-announcement quarters, quarter 3 and quarter 4 in 2010. Theresults are available upon request.

39Nutritional restrictions during the prenatal period are not necessarily reflected in lower birthweights: for example, individuals exposed in utero in early gestation to the Dutch famine did not presentlower birth weights but higher rate of incidence of coronary heart diseases, diabetes and obesity as com-pared to non-exposed individuals ( [Painter et al., 2005]; [Roseboom et al., 2011]).

29

the 3rd trimester to negative economic fluctuations in Argentina, and only for chil-dren of low educated mothers who were likely credit constrained. Yet, [Almondand Mazumder, 2011] look at relatively mild forms of nutritional disruptions im-posed by Ramadan daylight fasting during pregnancy and find a negative impacton birth weights, but only for children exposed during the first two trimesters ofpregnancy.

From this evidence, it is safe to conclude that possible nutritional restrictionssuffered by the fetus would lead to worsening (or unchanged) weight at birth,whereas we find improvements in birth weight. Additionally, we show in Table 2.1,column (4) that there were no significant change in foodstuff expenditures follow-ing the wage cut. Overall, we may safely conclude that the nutrition channel is notdriving our main results.

Health damaging goods. A decrease in household income may also inducea reduction in the consumption of health-damaging goods, such as cigarettes andalcohol. The medical literature shows that maternal smoking or alcohol consump-tion during pregnancy correlate with the increased risk of miscarriage and low birthweight ( [Floyd et al., 1993]). [Ruhm and Black, 2002] and [Ruhm, 2003] show thathealth-related behavioural improvements, in the form of decreased consumptionof alcohol and cigarettes, have a counter-cyclical pattern and the average healthlevel improves during recessions. [Dehejia and Lleras-Muney, 2004] find signifi-cant improvements in infant health outcomes at birth due to changes in individualbehaviour of white mothers who significantly reduced smoking and alcohol con-sumption during pregnancy. These behavioural improvements were sufficientlystrong to offset the simultaneous negative selection into motherhood.

Unfortunately, information on mothers smoking or drinking habits is not in-cluded in the VSN. Evidence from RHBS in Table 2.1 shows no change in alcoholand cigarettes expenditures per capita induced by the austerity measures. Theseexpenditures reflect the behaviour of the average individual/households and notpregnant women. Even if behavioural improvements did occur, we observe signif-icant changes for boys only, in their 1st trimester of pregnancy and, to our knowl-edge, it has not been determined that boys would benefit more than girls frombehavioural improvements (in early gestation). We argue the behavioural improve-ments of pregnant mothers is not likely to be the main channel through which theausterity measures influenced health at birth, though we can not dismiss its role.

30

[Nilsson, 2014] finds that boys exposed early in utero to an increase in theavailability of alcohol in Sweden were the most negatively affected at birth as mea-sured by a reduced share of males, which indicates that boys highly exposed toalcohol were more likely to be spontaneously aborted. If pregnant women reactedto the austerity-induced shock by increasing alcohol intake (especially before preg-nancy recognition), we may also explain our results through increased spontaneousabortions of the weakest male fetuses. We will investigate this in the next section.

Prenatal care and labor supply. A decrease in wage may also lower the op-portunity cost of leisure and health-improving activities (bed rest in high-risk preg-nancies), and may induce a shift in the labour supply of pregnant women from full-to part-time employment which would positively influence childrens outcomes atbirth. This is unlikely due to the rigidity of the public sector employment in Ro-mania and the limited opportunities of part time public employment: less than 1%of public sector employees have a part-time contract (source: RHBS). Women em-ployed in the private sector could have reacted to the significant wage cut by: 1)an increased rate of absenteeism, thus increasing their leisure time; the RHBS in-formation on absenteeism does not reveal any significant differences between 2010and 2007-2009 for women employed in the public sector; 2) changing occupa-tional status; RHBS reveals a very high degree of persistence in the occupationalstatus, with about 99% women having the same occupational status as in the last 12months (both for employed and housewife mothers);40 also, there is no change afterthe wage cut announcement in the share of housewives that used to be employedin the prior 12 months; 3) changing sector of employment; this channel seems un-likely since the unemployment rates in the public sector were high and rising, andthat employment rates in the public sector were stable over the entire period.

A wage cut may potentially restrict the antenatal medical supervision by low-ering the number of prenatal medical visits. In Romania, prenatal care is free ofcharge and is available to all pregnant women irrespective of their employment sta-tus, therefore it is unlikely that publicly employed mothers would reduce their useof prenatal care.

40We check if women potentially on the margin of leaving the labor force due to a problematic or afirst-child pregnancy are more likely to exit the labor force and become housewives after the wage cut.We test whether the number of housewife mothers significantly changes in 2010 for the first born chil-dren and for births that signal a problematic pregnancy: very preterm birth (before the 32nd gestationalweek) and very low birth weight (a birth weight less than 1,500 grams) and find no such effect.

31

2.4.3 Prenatal stress

An unexpected and significant income shock may induce psychological distressdue to the financial insecurity it entails. Indeed, 2010 survey evidence indicateshigher stress levels, particularly related to inadequate wages, among the staff in thepublic vs. the private Romanian sector ( [Spielberger et al., 2010]). The psycholog-ical stress caused by the austerity shock experienced by the pregnant women mayinfluence the fetal development through higher levels of cortisol, a stress hormonethat reaches the fetus. The exposure to high cortisol levels induces structural adap-tations in order to accelerate the maturation of the fetus and ensure her survival ina predicted stressful environment, but also to modify her ulterior response to stress( [Gluckman and Hanson, 2005]). Though these predictive adaptive responses arenot necessarily reflected in birth outcomes (but may manifest later), numerous med-ical studies have identified a direct link between prenatal stress exposure and in-creased incidence of preterm delivery and low birth weight or increased risk of amiscarriage (see [Mulder et al., 2002], [Maconochie et al., 2007], [Beydoun andSaftlas, 2008] for comprehensive reviews).

In addition to the medical literature, there is a growing interest among economiststo quantify the effects of maternal stress on infant birth outcomes by exploiting in-stances in which stress is generated by exogenous events. The evidence shows thatearly exposure to stress is more likely to harm a childs outcome at birth. [Cama-cho, 2008] finds a negative impact of stress induced by landmine explosions oninfant birth weight when exposure occurs during the 1st trimester of the pregnancy,while [Mansour and Rees, 2012] identify a causal relationship between the numberof fatalities in an armed conflict that occur during the 1st trimester of pregnancyand increased probability of low birth weight. [Bozzoli and Quintana-Domeque,2013] find increased low birth weight incidence due to negative macroeconomicfluctuations for children in the 1st trimester which they attribute to maternal stress,occurring both to high and low educated mothers. On the other hand, [Aizer et al.,2009] use cortisol levels during pregnancy in a mother fixed effects strategy andfinds no negative effects of maternal prenatal stress on health at birth, althoughthey find significant negative effects on other long term outcomes.

Selection in utero. The evidence so far suggests that prenatal stress scarssurvivors, leading to worse health outcomes at birth. However, medical evidence

32

shows that prenatal maternal stress could also lead to improved average healthoutcomes at birth by means of a natural selection mechanism, whereby prenatalmaternal stress raises the fitness criterion required to avoid spontaneous abortion.In particular, the theory of selection in utero hypothesises that weaker fetuses arespontaneously aborted because of significant maternal stress, and that the weakmale fetuses are being aborted more often than weak female fetuses. [Trivers andWillard, 1973] hypothesis postulates that the selection mechanism preponderantlyselects against weaker male fetuses, as the likelihood of reproductive success ofa weak male is relatively lower than that of a weak female. An alternative expla-nation for the more frequent miscarriage of males relative to females is related tomales more rapid growth rate during early pregnancy, which makes males morepredisposed to abnormalities than female fetuses and thus more exposed to riskof spontaneous abortion. Medical evidence indicates that selection in utero af-fects fetuses in their early developmental stages ( [Hobel et al., 1999], [Owen andMatthews, 2003], [Catalano et al., 2012]).

The selective mortality mechanism is reflected in a decrease of the sex-ratio atbirth and in the improvement of the average health level for the male cohort ex-posed in utero to the stressor. [Catalano et al., 2012] find an inverse relationshipbetween maternal cortisol levels during pregnancy and male cohort size and con-clude that elevated maternal stress culls cohorts by raising the fitness criterion,thus resulting in healthier males. [Catalano et al., 2010] show that mass layoffspredict lower secondary sex ratios as a consequence of significant maternal stressduring pregnancy due to adverse economic conditions that preponderantly selectsagainst weak male fetuses. [Sanders and Stoecker, 2011] evidnece that gender ratiosat birth can be used to infer fetal death rates of males, which are more vulnerableto maternal stress. Finally, [Valente, 2015] finds evidence of selection in utero dueto maternal conflict exposure. Our results so far indicate significant improvementsin health at birth of male cohorts exposed to the shock in early gestation.

2.5 Further evidence of selection in utero

The evidence from Section 2.4 seems to indicate that selection in utero, caused byin-utero maternal stress and/or increased smoking or alcohol intake, may help ex-plain our positive effect on health at birth for boys exposed to the shock startingvery early in the pregnancy. Because we do not have data on miscarriages, a com-

33

mon problem in the literature, we proceed to examine the effects on the secondarysex-ratio and the cohort size.

2.5.1 Sex-ratio at birth

Similar to other studies with individual level data, we model the sex-ratio at birth asthe probability of a male birth. Panel A of Table 2.8 presents the results of the DDestimation for the probability of a live birth being a male, using a similar frame-work as before, while Panel B show results for the high educated mothers. Theoverall effect on the probability of a child being a boy, in Panel A, for the childrenwho were in the 1st trimester of gestation at the time of the shock is negative andsignificant in columns (1) and (2) with a magnitude of about 3.3 pp (6.7 % of themean), and marginally significant (p-value=0.105) in column (3), when we includethe pre-treatment dynamics. This effect seems to be driven by the high-